SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/backing-dev.h>
  27 #include <linux/mpage.h>
  28 #include <linux/swap.h>
  29 #include <linux/writeback.h>
  30 #include <linux/statfs.h>
  31 #include <linux/compat.h>
  32 #include <linux/version.h>
  33 #include "ctree.h"
  34 #include "disk-io.h"
  35 #include "transaction.h"
  36 #include "btrfs_inode.h"
  37 #include "ioctl.h"
  38 #include "print-tree.h"
  39 #include "compat.h"
  40
  41
  42 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
  43                                 struct page **prepared_pages,
  44                                 const char __user * buf)
  45 {
  46         long page_fault = 0;
  47         int i;
  48         int offset = pos & (PAGE_CACHE_SIZE - 1);
  49
  50         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  51                 size_t count = min_t(size_t,
  52                                      PAGE_CACHE_SIZE - offset, write_bytes);
  53                 struct page *page = prepared_pages[i];
  54                 fault_in_pages_readable(buf, count);
  55
  56                 /* Copy data from userspace to the current page */
  57                 kmap(page);
  58                 page_fault = __copy_from_user(page_address(page) + offset,
  59                                               buf, count);
  60                 /* Flush processor's dcache for this page */
  61                 flush_dcache_page(page);
  62                 kunmap(page);
  63                 buf += count;
  64                 write_bytes -= count;
  65
  66                 if (page_fault)
  67                         break;
  68         }
  69         return page_fault ? -EFAULT : 0;
  70 }
  71
  72 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
  73 {
  74         size_t i;
  75         for (i = 0; i < num_pages; i++) {
  76                 if (!pages[i])
  77                         break;
  78                 ClearPageChecked(pages[i]);
  79                 unlock_page(pages[i]);
  80                 mark_page_accessed(pages[i]);
  81                 page_cache_release(pages[i]);
  82         }
  83 }
  84
  85 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
  86                                 struct btrfs_root *root, struct inode *inode,
  87                                 u64 offset, size_t size,
  88                                 struct page **pages, size_t page_offset,
  89                                 int num_pages)
  90 {
  91         struct btrfs_key key;
  92         struct btrfs_path *path;
  93         struct extent_buffer *leaf;
  94         char *kaddr;
  95         unsigned long ptr;
  96         struct btrfs_file_extent_item *ei;
  97         struct page *page;
  98         u32 datasize;
  99         int err = 0;
 100         int ret;
 101         int i;
 102         ssize_t cur_size;
 103
 104         path = btrfs_alloc_path();
 105         if (!path)
 106                 return -ENOMEM;
 107
 108         btrfs_set_trans_block_group(trans, inode);
 109
 110         key.objectid = inode->i_ino;
 111         key.offset = offset;
 112         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 113
 114         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 115         if (ret < 0) {
 116                 err = ret;
 117                 goto fail;
 118         }
 119         if (ret == 1) {
 120                 struct btrfs_key found_key;
 121
 122                 if (path->slots[0] == 0)
 123                         goto insert;
 124
 125                 path->slots[0]--;
 126                 leaf = path->nodes[0];
 127                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 128
 129                 if (found_key.objectid != inode->i_ino)
 130                         goto insert;
 131
 132                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 133                         goto insert;
 134                 ei = btrfs_item_ptr(leaf, path->slots[0],
 135                                     struct btrfs_file_extent_item);
 136
 137                 if (btrfs_file_extent_type(leaf, ei) !=
 138                     BTRFS_FILE_EXTENT_INLINE) {
 139                         goto insert;
 140                 }
 141                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 142                 ret = 0;
 143         }
 144         if (ret == 0) {
 145                 u32 found_size;
 146                 u64 found_end;
 147
 148                 leaf = path->nodes[0];
 149                 ei = btrfs_item_ptr(leaf, path->slots[0],
 150                                     struct btrfs_file_extent_item);
 151
 152                 if (btrfs_file_extent_type(leaf, ei) !=
 153                     BTRFS_FILE_EXTENT_INLINE) {
 154                         err = ret;
 155                         btrfs_print_leaf(root, leaf);
 156                         printk("found wasn't inline offset %Lu inode %lu\n",
 157                                offset, inode->i_ino);
 158                         goto fail;
 159                 }
 160                 found_size = btrfs_file_extent_inline_len(leaf,
 161                                           btrfs_item_nr(leaf, path->slots[0]));
 162                 found_end = key.offset + found_size;
 163
 164                 if (found_end < offset + size) {
 165                         btrfs_release_path(root, path);
 166                         ret = btrfs_search_slot(trans, root, &key, path,
 167                                                 offset + size - found_end, 1);
 168                         BUG_ON(ret != 0);
 169
 170                         ret = btrfs_extend_item(trans, root, path,
 171                                                 offset + size - found_end);
 172                         if (ret) {
 173                                 err = ret;
 174                                 goto fail;
 175                         }
 176                         leaf = path->nodes[0];
 177                         ei = btrfs_item_ptr(leaf, path->slots[0],
 178                                             struct btrfs_file_extent_item);
 179                         inode->i_blocks += (offset + size - found_end) >> 9;
 180                 }
 181                 if (found_end < offset) {
 182                         ptr = btrfs_file_extent_inline_start(ei) + found_size;
 183                         memset_extent_buffer(leaf, 0, ptr, offset - found_end);
 184                 }
 185         } else {
 186 insert:
 187                 btrfs_release_path(root, path);
 188                 datasize = offset + size - key.offset;
 189                 inode->i_blocks += datasize >> 9;
 190                 datasize = btrfs_file_extent_calc_inline_size(datasize);
 191                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 192                                               datasize);
 193                 if (ret) {
 194                         err = ret;
 195                         printk("got bad ret %d\n", ret);
 196                         goto fail;
 197                 }
 198                 leaf = path->nodes[0];
 199                 ei = btrfs_item_ptr(leaf, path->slots[0],
 200                                     struct btrfs_file_extent_item);
 201                 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 202                 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 203         }
 204         ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
 205
 206         cur_size = size;
 207         i = 0;
 208         while (size > 0) {
 209                 page = pages[i];
 210                 kaddr = kmap_atomic(page, KM_USER0);
 211                 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
 212                 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
 213                 kunmap_atomic(kaddr, KM_USER0);
 214                 page_offset = 0;
 215                 ptr += cur_size;
 216                 size -= cur_size;
 217                 if (i >= num_pages) {
 218                         printk("i %d num_pages %d\n", i, num_pages);
 219                 }
 220                 i++;
 221         }
 222         btrfs_mark_buffer_dirty(leaf);
 223 fail:
 224         btrfs_free_path(path);
 225         return err;
 226 }
 227
 228 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 229                                    struct btrfs_root *root,
 230                                    struct file *file,
 231                                    struct page **pages,
 232                                    size_t num_pages,
 233                                    loff_t pos,
 234                                    size_t write_bytes)
 235 {
 236         int err = 0;
 237         int i;
 238         struct inode *inode = fdentry(file)->d_inode;
 239         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 240         u64 hint_byte;
 241         u64 num_bytes;
 242         u64 start_pos;
 243         u64 end_of_last_block;
 244         u64 end_pos = pos + write_bytes;
 245         u64 inline_size;
 246         int did_inline = 0;
 247         loff_t isize = i_size_read(inode);
 248
 249         start_pos = pos & ~((u64)root->sectorsize - 1);
 250         num_bytes = (write_bytes + pos - start_pos +
 251                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 252
 253         end_of_last_block = start_pos + num_bytes - 1;
 254
 255         lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 256         trans = btrfs_join_transaction(root, 1);
 257         if (!trans) {
 258                 err = -ENOMEM;
 259                 goto out_unlock;
 260         }
 261         btrfs_set_trans_block_group(trans, inode);
 262         hint_byte = 0;
 263
 264         if ((end_of_last_block & 4095) == 0) {
 265                 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
 266         }
 267         set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 268
 269         /* FIXME...EIEIO, ENOSPC and more */
 270         /* insert any holes we need to create */
 271         if (isize < start_pos) {
 272                 u64 last_pos_in_file;
 273                 u64 hole_size;
 274                 u64 mask = root->sectorsize - 1;
 275                 last_pos_in_file = (isize + mask) & ~mask;
 276                 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
 277                 if (hole_size > 0) {
 278                         btrfs_wait_ordered_range(inode, last_pos_in_file,
 279                                                  last_pos_in_file + hole_size);
 280                         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 281                         err = btrfs_drop_extents(trans, root, inode,
 282                                                  last_pos_in_file,
 283                                                  last_pos_in_file + hole_size,
 284                                                  last_pos_in_file,
 285                                                  &hint_byte);
 286                         if (err)
 287                                 goto failed;
 288
 289                         err = btrfs_insert_file_extent(trans, root,
 290                                                        inode->i_ino,
 291                                                        last_pos_in_file,
 292                                                        0, 0, hole_size, 0);
 293                         btrfs_drop_extent_cache(inode, last_pos_in_file,
 294                                         last_pos_in_file + hole_size -1);
 295                         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 296                         btrfs_check_file(root, inode);
 297                 }
 298                 if (err)
 299                         goto failed;
 300         }
 301
 302         /*
 303          * either allocate an extent for the new bytes or setup the key
 304          * to show we are doing inline data in the extent
 305          */
 306         inline_size = end_pos;
 307         if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 308             inline_size > root->fs_info->max_inline ||
 309             (inline_size & (root->sectorsize -1)) == 0 ||
 310             inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 311                 /* check for reserved extents on each page, we don't want
 312                  * to reset the delalloc bit on things that already have
 313                  * extents reserved.
 314                  */
 315                 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 316                 for (i = 0; i < num_pages; i++) {
 317                         struct page *p = pages[i];
 318                         SetPageUptodate(p);
 319                         ClearPageChecked(p);
 320                         set_page_dirty(p);
 321                 }
 322         } else {
 323                 u64 aligned_end;
 324                 /* step one, delete the existing extents in this range */
 325                 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 326                         ~((u64)root->sectorsize - 1);
 327                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 328                 err = btrfs_drop_extents(trans, root, inode, start_pos,
 329                                          aligned_end, aligned_end, &hint_byte);
 330                 if (err)
 331                         goto failed;
 332                 if (isize > inline_size)
 333                         inline_size = min_t(u64, isize, aligned_end);
 334                 inline_size -= start_pos;
 335                 err = insert_inline_extent(trans, root, inode, start_pos,
 336                                            inline_size, pages, 0, num_pages);
 337                 btrfs_drop_extent_cache(inode, start_pos, aligned_end - 1);
 338                 BUG_ON(err);
 339                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 340
 341                 /*
 342                  * an ugly way to do all the prop accounting around
 343                  * the page bits and mapping tags
 344                  */
 345                 set_page_writeback(pages[0]);
 346                 end_page_writeback(pages[0]);
 347                 did_inline = 1;
 348         }
 349         if (end_pos > isize) {
 350                 i_size_write(inode, end_pos);
 351                 if (did_inline)
 352                         BTRFS_I(inode)->disk_i_size = end_pos;
 353                 btrfs_update_inode(trans, root, inode);
 354         }
 355 failed:
 356         err = btrfs_end_transaction(trans, root);
 357 out_unlock:
 358         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
 359         return err;
 360 }
 361
 362 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
 363 {
 364         struct extent_map *em;
 365         struct extent_map *split = NULL;
 366         struct extent_map *split2 = NULL;
 367         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 368         u64 len = end - start + 1;
 369         int ret;
 370         int testend = 1;
 371
 372         WARN_ON(end < start);
 373         if (end == (u64)-1) {
 374                 len = (u64)-1;
 375                 testend = 0;
 376         }
 377         while(1) {
 378                 if (!split)
 379                         split = alloc_extent_map(GFP_NOFS);
 380                 if (!split2)
 381                         split2 = alloc_extent_map(GFP_NOFS);
 382
 383                 spin_lock(&em_tree->lock);
 384                 em = lookup_extent_mapping(em_tree, start, len);
 385                 if (!em) {
 386                         spin_unlock(&em_tree->lock);
 387                         break;
 388                 }
 389                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 390                 remove_extent_mapping(em_tree, em);
 391
 392                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 393                     em->start < start) {
 394                         split->start = em->start;
 395                         split->len = start - em->start;
 396                         split->block_start = em->block_start;
 397                         split->bdev = em->bdev;
 398                         split->flags = em->flags;
 399                         ret = add_extent_mapping(em_tree, split);
 400                         BUG_ON(ret);
 401                         free_extent_map(split);
 402                         split = split2;
 403                         split2 = NULL;
 404                 }
 405                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 406                     testend && em->start + em->len > start + len) {
 407                         u64 diff = start + len - em->start;
 408
 409                         split->start = start + len;
 410                         split->len = em->start + em->len - (start + len);
 411                         split->bdev = em->bdev;
 412                         split->flags = em->flags;
 413
 414                         split->block_start = em->block_start + diff;
 415
 416                         ret = add_extent_mapping(em_tree, split);
 417                         BUG_ON(ret);
 418                         free_extent_map(split);
 419                         split = NULL;
 420                 }
 421                 spin_unlock(&em_tree->lock);
 422
 423                 /* once for us */
 424                 free_extent_map(em);
 425                 /* once for the tree*/
 426                 free_extent_map(em);
 427         }
 428         if (split)
 429                 free_extent_map(split);
 430         if (split2)
 431                 free_extent_map(split2);
 432         return 0;
 433 }
 434
 435 int btrfs_check_file(struct btrfs_root *root, struct inode *inode)
 436 {
 437         return 0;
 438 #if 0
 439         struct btrfs_path *path;
 440         struct btrfs_key found_key;
 441         struct extent_buffer *leaf;
 442         struct btrfs_file_extent_item *extent;
 443         u64 last_offset = 0;
 444         int nritems;
 445         int slot;
 446         int found_type;
 447         int ret;
 448         int err = 0;
 449         u64 extent_end = 0;
 450
 451         path = btrfs_alloc_path();
 452         ret = btrfs_lookup_file_extent(NULL, root, path, inode->i_ino,
 453                                        last_offset, 0);
 454         while(1) {
 455                 nritems = btrfs_header_nritems(path->nodes[0]);
 456                 if (path->slots[0] >= nritems) {
 457                         ret = btrfs_next_leaf(root, path);
 458                         if (ret)
 459                                 goto out;
 460                         nritems = btrfs_header_nritems(path->nodes[0]);
 461                 }
 462                 slot = path->slots[0];
 463                 leaf = path->nodes[0];
 464                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
 465                 if (found_key.objectid != inode->i_ino)
 466                         break;
 467                 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
 468                         goto out;
 469
 470                 if (found_key.offset < last_offset) {
 471                         WARN_ON(1);
 472                         btrfs_print_leaf(root, leaf);
 473                         printk("inode %lu found offset %Lu expected %Lu\n",
 474                                inode->i_ino, found_key.offset, last_offset);
 475                         err = 1;
 476                         goto out;
 477                 }
 478                 extent = btrfs_item_ptr(leaf, slot,
 479                                         struct btrfs_file_extent_item);
 480                 found_type = btrfs_file_extent_type(leaf, extent);
 481                 if (found_type == BTRFS_FILE_EXTENT_REG) {
 482                         extent_end = found_key.offset +
 483                              btrfs_file_extent_num_bytes(leaf, extent);
 484                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 485                         struct btrfs_item *item;
 486                         item = btrfs_item_nr(leaf, slot);
 487                         extent_end = found_key.offset +
 488                              btrfs_file_extent_inline_len(leaf, item);
 489                         extent_end = (extent_end + root->sectorsize - 1) &
 490                                 ~((u64)root->sectorsize -1 );
 491                 }
 492                 last_offset = extent_end;
 493                 path->slots[0]++;
 494         }
 495         if (0 && last_offset < inode->i_size) {
 496                 WARN_ON(1);
 497                 btrfs_print_leaf(root, leaf);
 498                 printk("inode %lu found offset %Lu size %Lu\n", inode->i_ino,
 499                        last_offset, inode->i_size);
 500                 err = 1;
 501
 502         }
 503 out:
 504         btrfs_free_path(path);
 505         return err;
 506 #endif
 507 }
 508
 509 /*
 510  * this is very complex, but the basic idea is to drop all extents
 511  * in the range start - end.  hint_block is filled in with a block number
 512  * that would be a good hint to the block allocator for this file.
 513  *
 514  * If an extent intersects the range but is not entirely inside the range
 515  * it is either truncated or split.  Anything entirely inside the range
 516  * is deleted from the tree.
 517  */
 518 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 519                        struct btrfs_root *root, struct inode *inode,
 520                        u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 521 {
 522         u64 extent_end = 0;
 523         u64 search_start = start;
 524         struct extent_buffer *leaf;
 525         struct btrfs_file_extent_item *extent;
 526         struct btrfs_path *path;
 527         struct btrfs_key key;
 528         struct btrfs_file_extent_item old;
 529         int keep;
 530         int slot;
 531         int bookend;
 532         int found_type;
 533         int found_extent;
 534         int found_inline;
 535         int recow;
 536         int ret;
 537
 538         btrfs_drop_extent_cache(inode, start, end - 1);
 539
 540         path = btrfs_alloc_path();
 541         if (!path)
 542                 return -ENOMEM;
 543         while(1) {
 544                 recow = 0;
 545                 btrfs_release_path(root, path);
 546                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 547                                                search_start, -1);
 548                 if (ret < 0)
 549                         goto out;
 550                 if (ret > 0) {
 551                         if (path->slots[0] == 0) {
 552                                 ret = 0;
 553                                 goto out;
 554                         }
 555                         path->slots[0]--;
 556                 }
 557 next_slot:
 558                 keep = 0;
 559                 bookend = 0;
 560                 found_extent = 0;
 561                 found_inline = 0;
 562                 extent = NULL;
 563                 leaf = path->nodes[0];
 564                 slot = path->slots[0];
 565                 ret = 0;
 566                 btrfs_item_key_to_cpu(leaf, &key, slot);
 567                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
 568                     key.offset >= end) {
 569                         goto out;
 570                 }
 571                 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
 572                     key.objectid != inode->i_ino) {
 573                         goto out;
 574                 }
 575                 if (recow) {
 576                         search_start = key.offset;
 577                         continue;
 578                 }
 579                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
 580                         extent = btrfs_item_ptr(leaf, slot,
 581                                                 struct btrfs_file_extent_item);
 582                         found_type = btrfs_file_extent_type(leaf, extent);
 583                         if (found_type == BTRFS_FILE_EXTENT_REG) {
 584                                 extent_end =
 585                                      btrfs_file_extent_disk_bytenr(leaf,
 586                                                                    extent);
 587                                 if (extent_end)
 588                                         *hint_byte = extent_end;
 589
 590                                 extent_end = key.offset +
 591                                      btrfs_file_extent_num_bytes(leaf, extent);
 592                                 found_extent = 1;
 593                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 594                                 struct btrfs_item *item;
 595                                 item = btrfs_item_nr(leaf, slot);
 596                                 found_inline = 1;
 597                                 extent_end = key.offset +
 598                                      btrfs_file_extent_inline_len(leaf, item);
 599                         }
 600                 } else {
 601                         extent_end = search_start;
 602                 }
 603
 604                 /* we found nothing we can drop */
 605                 if ((!found_extent && !found_inline) ||
 606                     search_start >= extent_end) {
 607                         int nextret;
 608                         u32 nritems;
 609                         nritems = btrfs_header_nritems(leaf);
 610                         if (slot >= nritems - 1) {
 611                                 nextret = btrfs_next_leaf(root, path);
 612                                 if (nextret)
 613                                         goto out;
 614                                 recow = 1;
 615                         } else {
 616                                 path->slots[0]++;
 617                         }
 618                         goto next_slot;
 619                 }
 620
 621                 if (found_inline) {
 622                         u64 mask = root->sectorsize - 1;
 623                         search_start = (extent_end + mask) & ~mask;
 624                 } else
 625                         search_start = extent_end;
 626                 if (end <= extent_end && start >= key.offset && found_inline) {
 627                         *hint_byte = EXTENT_MAP_INLINE;
 628                         continue;
 629                 }
 630                 if (end < extent_end && end >= key.offset) {
 631                         if (found_extent) {
 632                                 u64 disk_bytenr =
 633                                     btrfs_file_extent_disk_bytenr(leaf, extent);
 634                                 u64 disk_num_bytes =
 635                                     btrfs_file_extent_disk_num_bytes(leaf,
 636                                                                       extent);
 637                                 read_extent_buffer(leaf, &old,
 638                                                    (unsigned long)extent,
 639                                                    sizeof(old));
 640                                 if (disk_bytenr != 0) {
 641                                         ret = btrfs_inc_extent_ref(trans, root,
 642                                                  disk_bytenr, disk_num_bytes,
 643                                                  root->root_key.objectid,
 644                                                  trans->transid,
 645                                                  key.objectid, end);
 646                                         BUG_ON(ret);
 647                                 }
 648                         }
 649                         bookend = 1;
 650                         if (found_inline && start <= key.offset)
 651                                 keep = 1;
 652                 }
 653                 /* truncate existing extent */
 654                 if (start > key.offset) {
 655                         u64 new_num;
 656                         u64 old_num;
 657                         keep = 1;
 658                         WARN_ON(start & (root->sectorsize - 1));
 659                         if (found_extent) {
 660                                 new_num = start - key.offset;
 661                                 old_num = btrfs_file_extent_num_bytes(leaf,
 662                                                                       extent);
 663                                 *hint_byte =
 664                                         btrfs_file_extent_disk_bytenr(leaf,
 665                                                                       extent);
 666                                 if (btrfs_file_extent_disk_bytenr(leaf,
 667                                                                   extent)) {
 668                                         dec_i_blocks(inode, old_num - new_num);
 669                                 }
 670                                 btrfs_set_file_extent_num_bytes(leaf, extent,
 671                                                                 new_num);
 672                                 btrfs_mark_buffer_dirty(leaf);
 673                         } else if (key.offset < inline_limit &&
 674                                    (end > extent_end) &&
 675                                    (inline_limit < extent_end)) {
 676                                 u32 new_size;
 677                                 new_size = btrfs_file_extent_calc_inline_size(
 678                                                    inline_limit - key.offset);
 679                                 dec_i_blocks(inode, (extent_end - key.offset) -
 680                                         (inline_limit - key.offset));
 681                                 btrfs_truncate_item(trans, root, path,
 682                                                     new_size, 1);
 683                         }
 684                 }
 685                 /* delete the entire extent */
 686                 if (!keep) {
 687                         u64 disk_bytenr = 0;
 688                         u64 disk_num_bytes = 0;
 689                         u64 extent_num_bytes = 0;
 690                         u64 root_gen;
 691                         u64 root_owner;
 692
 693                         root_gen = btrfs_header_generation(leaf);
 694                         root_owner = btrfs_header_owner(leaf);
 695                         if (found_extent) {
 696                                 disk_bytenr =
 697                                       btrfs_file_extent_disk_bytenr(leaf,
 698                                                                      extent);
 699                                 disk_num_bytes =
 700                                       btrfs_file_extent_disk_num_bytes(leaf,
 701                                                                        extent);
 702                                 extent_num_bytes =
 703                                       btrfs_file_extent_num_bytes(leaf, extent);
 704                                 *hint_byte =
 705                                         btrfs_file_extent_disk_bytenr(leaf,
 706                                                                       extent);
 707                         }
 708                         ret = btrfs_del_item(trans, root, path);
 709                         /* TODO update progress marker and return */
 710                         BUG_ON(ret);
 711                         btrfs_release_path(root, path);
 712                         extent = NULL;
 713                         if (found_extent && disk_bytenr != 0) {
 714                                 dec_i_blocks(inode, extent_num_bytes);
 715                                 ret = btrfs_free_extent(trans, root,
 716                                                 disk_bytenr,
 717                                                 disk_num_bytes,
 718                                                 root_owner,
 719                                                 root_gen, inode->i_ino,
 720                                                 key.offset, 0);
 721                         }
 722
 723                         BUG_ON(ret);
 724                         if (!bookend && search_start >= end) {
 725                                 ret = 0;
 726                                 goto out;
 727                         }
 728                         if (!bookend)
 729                                 continue;
 730                 }
 731                 if (bookend && found_inline && start <= key.offset) {
 732                         u32 new_size;
 733                         new_size = btrfs_file_extent_calc_inline_size(
 734                                                    extent_end - end);
 735                         dec_i_blocks(inode, (extent_end - key.offset) -
 736                                         (extent_end - end));
 737                         btrfs_truncate_item(trans, root, path, new_size, 0);
 738                 }
 739                 /* create bookend, splitting the extent in two */
 740                 if (bookend && found_extent) {
 741                         struct btrfs_key ins;
 742                         ins.objectid = inode->i_ino;
 743                         ins.offset = end;
 744                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
 745                         btrfs_release_path(root, path);
 746                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
 747                                                       sizeof(*extent));
 748
 749                         leaf = path->nodes[0];
 750                         if (ret) {
 751                                 btrfs_print_leaf(root, leaf);
 752                                 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.type, ins.offset, start, end, key.offset, extent_end, keep);
 753                         }
 754                         BUG_ON(ret);
 755                         extent = btrfs_item_ptr(leaf, path->slots[0],
 756                                                 struct btrfs_file_extent_item);
 757                         write_extent_buffer(leaf, &old,
 758                                             (unsigned long)extent, sizeof(old));
 759
 760                         btrfs_set_file_extent_offset(leaf, extent,
 761                                     le64_to_cpu(old.offset) + end - key.offset);
 762                         WARN_ON(le64_to_cpu(old.num_bytes) <
 763                                 (extent_end - end));
 764                         btrfs_set_file_extent_num_bytes(leaf, extent,
 765                                                         extent_end - end);
 766                         btrfs_set_file_extent_type(leaf, extent,
 767                                                    BTRFS_FILE_EXTENT_REG);
 768
 769                         btrfs_mark_buffer_dirty(path->nodes[0]);
 770                         if (le64_to_cpu(old.disk_bytenr) != 0) {
 771                                 inode->i_blocks +=
 772                                       btrfs_file_extent_num_bytes(leaf,
 773                                                                   extent) >> 9;
 774                         }
 775                         ret = 0;
 776                         goto out;
 777                 }
 778         }
 779 out:
 780         btrfs_free_path(path);
 781         btrfs_check_file(root, inode);
 782         return ret;
 783 }
 784
 785 /*
 786  * this gets pages into the page cache and locks them down
 787  */
 788 static int prepare_pages(struct btrfs_root *root, struct file *file,
 789                          struct page **pages, size_t num_pages,
 790                          loff_t pos, unsigned long first_index,
 791                          unsigned long last_index, size_t write_bytes)
 792 {
 793         int i;
 794         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 795         struct inode *inode = fdentry(file)->d_inode;
 796         int err = 0;
 797         u64 start_pos;
 798         u64 last_pos;
 799
 800         start_pos = pos & ~((u64)root->sectorsize - 1);
 801         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 802
 803         memset(pages, 0, num_pages * sizeof(struct page *));
 804 again:
 805         for (i = 0; i < num_pages; i++) {
 806                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 807                 if (!pages[i]) {
 808                         err = -ENOMEM;
 809                         BUG_ON(1);
 810                 }
 811                 wait_on_page_writeback(pages[i]);
 812         }
 813         if (start_pos < inode->i_size) {
 814                 struct btrfs_ordered_extent *ordered;
 815                 lock_extent(&BTRFS_I(inode)->io_tree,
 816                             start_pos, last_pos - 1, GFP_NOFS);
 817                 ordered = btrfs_lookup_first_ordered_extent(inode, last_pos -1);
 818                 if (ordered &&
 819                     ordered->file_offset + ordered->len > start_pos &&
 820                     ordered->file_offset < last_pos) {
 821                         btrfs_put_ordered_extent(ordered);
 822                         unlock_extent(&BTRFS_I(inode)->io_tree,
 823                                       start_pos, last_pos - 1, GFP_NOFS);
 824                         for (i = 0; i < num_pages; i++) {
 825                                 unlock_page(pages[i]);
 826                                 page_cache_release(pages[i]);
 827                         }
 828                         btrfs_wait_ordered_range(inode, start_pos,
 829                                                  last_pos - start_pos);
 830                         goto again;
 831                 }
 832                 if (ordered)
 833                         btrfs_put_ordered_extent(ordered);
 834
 835                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 836                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
 837                                   GFP_NOFS);
 838                 unlock_extent(&BTRFS_I(inode)->io_tree,
 839                               start_pos, last_pos - 1, GFP_NOFS);
 840         }
 841         for (i = 0; i < num_pages; i++) {
 842                 clear_page_dirty_for_io(pages[i]);
 843                 set_page_extent_mapped(pages[i]);
 844                 WARN_ON(!PageLocked(pages[i]));
 845         }
 846         return 0;
 847 }
 848
 849 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 850                                 size_t count, loff_t *ppos)
 851 {
 852         loff_t pos;
 853         loff_t start_pos;
 854         ssize_t num_written = 0;
 855         ssize_t err = 0;
 856         int ret = 0;
 857         struct inode *inode = fdentry(file)->d_inode;
 858         struct btrfs_root *root = BTRFS_I(inode)->root;
 859         struct page **pages = NULL;
 860         int nrptrs;
 861         struct page *pinned[2];
 862         unsigned long first_index;
 863         unsigned long last_index;
 864
 865         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 866                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 867         pinned[0] = NULL;
 868         pinned[1] = NULL;
 869
 870         pos = *ppos;
 871         start_pos = pos;
 872
 873         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 874         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 875         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 876         if (err)
 877                 goto out_nolock;
 878         if (count == 0)
 879                 goto out_nolock;
 880 #ifdef REMOVE_SUID_PATH
 881         err = remove_suid(&file->f_path);
 882 #else
 883 # if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
 884         err = file_remove_suid(file);
 885 # else
 886         err = remove_suid(fdentry(file));
 887 # endif
 888 #endif
 889         if (err)
 890                 goto out_nolock;
 891         file_update_time(file);
 892
 893         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 894
 895         mutex_lock(&inode->i_mutex);
 896         first_index = pos >> PAGE_CACHE_SHIFT;
 897         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 898
 899         /*
 900          * if this is a nodatasum mount, force summing off for the inode
 901          * all the time.  That way a later mount with summing on won't
 902          * get confused
 903          */
 904         if (btrfs_test_opt(root, NODATASUM))
 905                 btrfs_set_flag(inode, NODATASUM);
 906
 907         /*
 908          * there are lots of better ways to do this, but this code
 909          * makes sure the first and last page in the file range are
 910          * up to date and ready for cow
 911          */
 912         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 913                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 914                 if (!PageUptodate(pinned[0])) {
 915                         ret = btrfs_readpage(NULL, pinned[0]);
 916                         BUG_ON(ret);
 917                         wait_on_page_locked(pinned[0]);
 918                 } else {
 919                         unlock_page(pinned[0]);
 920                 }
 921         }
 922         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 923                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 924                 if (!PageUptodate(pinned[1])) {
 925                         ret = btrfs_readpage(NULL, pinned[1]);
 926                         BUG_ON(ret);
 927                         wait_on_page_locked(pinned[1]);
 928                 } else {
 929                         unlock_page(pinned[1]);
 930                 }
 931         }
 932
 933         while(count > 0) {
 934                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 935                 size_t write_bytes = min(count, nrptrs *
 936                                         (size_t)PAGE_CACHE_SIZE -
 937                                          offset);
 938                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 939                                         PAGE_CACHE_SHIFT;
 940
 941                 WARN_ON(num_pages > nrptrs);
 942                 memset(pages, 0, sizeof(pages));
 943
 944                 ret = btrfs_check_free_space(root, write_bytes, 0);
 945                 if (ret)
 946                         goto out;
 947
 948                 ret = prepare_pages(root, file, pages, num_pages,
 949                                     pos, first_index, last_index,
 950                                     write_bytes);
 951                 if (ret)
 952                         goto out;
 953
 954                 ret = btrfs_copy_from_user(pos, num_pages,
 955                                            write_bytes, pages, buf);
 956                 if (ret) {
 957                         btrfs_drop_pages(pages, num_pages);
 958                         goto out;
 959                 }
 960
 961                 ret = dirty_and_release_pages(NULL, root, file, pages,
 962                                               num_pages, pos, write_bytes);
 963                 btrfs_drop_pages(pages, num_pages);
 964                 if (ret)
 965                         goto out;
 966
 967                 buf += write_bytes;
 968                 count -= write_bytes;
 969                 pos += write_bytes;
 970                 num_written += write_bytes;
 971
 972                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
 973                 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 974                         btrfs_btree_balance_dirty(root, 1);
 975                 btrfs_throttle(root);
 976                 cond_resched();
 977         }
 978 out:
 979         mutex_unlock(&inode->i_mutex);
 980
 981 out_nolock:
 982         kfree(pages);
 983         if (pinned[0])
 984                 page_cache_release(pinned[0]);
 985         if (pinned[1])
 986                 page_cache_release(pinned[1]);
 987         *ppos = pos;
 988
 989         if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
 990                 err = sync_page_range(inode, inode->i_mapping,
 991                                       start_pos, num_written);
 992                 if (err < 0)
 993                         num_written = err;
 994         } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
 995 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
 996                 do_sync_file_range(file, start_pos,
 997                                       start_pos + num_written - 1,
 998                                       SYNC_FILE_RANGE_WRITE |
 999                                       SYNC_FILE_RANGE_WAIT_AFTER);
1000 #else
1001                 do_sync_mapping_range(inode->i_mapping, start_pos,
1002                                       start_pos + num_written - 1,
1003                                       SYNC_FILE_RANGE_WRITE |
1004                                       SYNC_FILE_RANGE_WAIT_AFTER);
1005 #endif
1006                 invalidate_mapping_pages(inode->i_mapping,
1007                       start_pos >> PAGE_CACHE_SHIFT,
1008                      (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1009         }
1010         current->backing_dev_info = NULL;
1011         return num_written ? num_written : err;
1012 }
1013
1014 int btrfs_release_file(struct inode * inode, struct file * filp)
1015 {
1016         if (filp->private_data)
1017                 btrfs_ioctl_trans_end(filp);
1018         return 0;
1019 }
1020
1021 static int btrfs_sync_file(struct file *file,
1022                            struct dentry *dentry, int datasync)
1023 {
1024         struct inode *inode = dentry->d_inode;
1025         struct btrfs_root *root = BTRFS_I(inode)->root;
1026         int ret = 0;
1027         struct btrfs_trans_handle *trans;
1028
1029         /*
1030          * check the transaction that last modified this inode
1031          * and see if its already been committed
1032          */
1033         if (!BTRFS_I(inode)->last_trans)
1034                 goto out;
1035
1036         mutex_lock(&root->fs_info->trans_mutex);
1037         if (BTRFS_I(inode)->last_trans <=
1038             root->fs_info->last_trans_committed) {
1039                 BTRFS_I(inode)->last_trans = 0;
1040                 mutex_unlock(&root->fs_info->trans_mutex);
1041                 goto out;
1042         }
1043         mutex_unlock(&root->fs_info->trans_mutex);
1044
1045         /*
1046          * ok we haven't committed the transaction yet, lets do a commit
1047          */
1048         if (file->private_data)
1049                 btrfs_ioctl_trans_end(file);
1050
1051         trans = btrfs_start_transaction(root, 1);
1052         if (!trans) {
1053                 ret = -ENOMEM;
1054                 goto out;
1055         }
1056         ret = btrfs_commit_transaction(trans, root);
1057 out:
1058         return ret > 0 ? EIO : ret;
1059 }
1060
1061 static struct vm_operations_struct btrfs_file_vm_ops = {
1062 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
1063         .nopage         = filemap_nopage,
1064         .populate       = filemap_populate,
1065 #else
1066         .fault          = filemap_fault,
1067 #endif
1068         .page_mkwrite   = btrfs_page_mkwrite,
1069 };
1070
1071 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1072 {
1073         vma->vm_ops = &btrfs_file_vm_ops;
1074         file_accessed(filp);
1075         return 0;
1076 }
1077
1078 struct file_operations btrfs_file_operations = {
1079         .llseek         = generic_file_llseek,
1080         .read           = do_sync_read,
1081         .aio_read       = generic_file_aio_read,
1082         .splice_read    = generic_file_splice_read,
1083 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1084         .sendfile       = generic_file_sendfile,
1085 #endif
1086         .write          = btrfs_file_write,
1087         .mmap           = btrfs_file_mmap,
1088         .open           = generic_file_open,
1089         .release        = btrfs_release_file,
1090         .fsync          = btrfs_sync_file,
1091         .unlocked_ioctl = btrfs_ioctl,
1092 #ifdef CONFIG_COMPAT
1093         .compat_ioctl   = btrfs_ioctl,
1094 #endif
1095 };